easyocr reproductibility of english model

[Shiro-LK]

Hello,

Thank you for your work. It is a very nice opensource framework.

I would like to reproduce (or at least having close results) to your english OCR model. Unfortunately, I haven't been able to get good r esults as soon as I am trying to predict multiple words at once. I suspect some parameters during my training is wrong so I have some question in order to reproduce your results as close as possible :

• does this file has the exact same parameters than the current model during training ? https://github.com/JaidedAI/EasyOCR/blob/master/trainer/config_files/en_filtered_config.yaml
• how did you create the generated dataset for the training ? https://github.com/Belval/TextRecognitionDataGenerator/tree/master/trdg On this repo it is possible to create it with different fonts, image background, color of the characters, width etc. Is it possible to provide these information and possibly the font/image background you use and maybe the text used for the generation ?

Thank for your help! @rkcosmos